Tracking Control of Optimal Profiles in a Nonlinear ... - ACS Publications

Jul 19, 2018 - The problem of optimal profiles tracking control under uncertainties for a ... 2017, 56, 6043], this work aims to improve the control s...
1 downloads 0 Views 4MB Size
Subscriber access provided by READING UNIV

Process Systems Engineering

Tracking Control of Optimal Profiles in a Nonlinear Fed-Batch Bioprocess under Parametric Uncertainty and Process Disturbances. María Nadia Pantano, María Cecilia Fernandez, Mario Emanuel Serrano, Oscar Alberto Ortiz, and Gustavo J. E. Scaglia Ind. Eng. Chem. Res., Just Accepted Manuscript • DOI: 10.1021/acs.iecr.8b01791 • Publication Date (Web): 19 Jul 2018 Downloaded from http://pubs.acs.org on July 27, 2018

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Tracking Control of Optimal Proles in a Nonlinear Fed-Batch Bioprocess under parametric uncertainty and process disturbances. María N. Pantano,



María C. Fernández, Mario E. Serrano, Oscar A. Ortiz, and

Gustavo J. E. Scaglia

Instituto de Ingeniería Química, Universidad Nacional de San Juan (UNSJ), CONICET, Av. Libertador San Martín (O) 1109, San Juan J5400ARL, Argentina E-mail: [email protected]

Abstract The problem of optimal proles tracking control under uncertainties for a nonlinear fed-batch bioprocess is addressed in this paper. Based on the results reported in, 1 this work aims to improve the control system response against parametric uncertainty and process disturbances. The methodology is simple, easy to implement, but with excellent results. The design parameters are optimized by a randomized Montecarlo algorithm. Besides, demonstration of the tracking error convergence to zero when the system is subjected to uncertainties is included in the article. The control system performance is tested through simulations, showing the improvement achieved.

1

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 36

Introduction Bioreactors performance is not only determined by productivity but also by process quality, which is mainly aected by the disturbances in the process variables. Therefore, nding a way to control these distortions is the main task to guarantee quality.

2

During the last decade, many researchers further investigated several control techniques applied to dierent bioprocesses,

37

especially after the implementation of Quality by De-

sign for biopharmaceuticals by the US Food and Drug Administration (FDA).

8

Particularly,

in fed-batch bioreactors many strategies were studied to improve the eciency and reproducibility of bioprocesses.

914

However, the problem that arises is the gap between scientic

research and the industry requirements. For example, usually research works on optimization and control strategies rarely consider model uncertainties, which are unavoidable in industrial processes,

1517

this could lead to a poor real life representation, and consequently,

to a bad performance with severe risks.

1820

Unfortunately, the high degree of non-linearity of the bioprocesses and the lack of highquality experimental data make modeling the system a challenge. Therefore, the resulting model presents a high degree of uncertainty. As a consequence, there are several mathematical models for the same biological process, including dierent structures and parameters but concordant with the available information of said processes. Uncertainties are one of the main obstacles in the development of advanced controllers for high-accuracy trajectory tracking control.

21

Not surprisingly, therefore, parametric un-

certainty has remained high up on the agenda of unsolved problems in control for the past three decades.

22

The main types of uncertainties that can be considered in a bioprocess are

the uncertain values of parameters (most frequent), time-varying model parameters, uncertain nonlinearities and unmodeled dynamics, among others. Besides that, there are external disturbances that are not modeled and aect the process variables too.

2325

Many existing works focus on the model uncertainty quantication due to the lack of knowledge of model parameters and the underlying physics by combining the results from

2

ACS Paragon Plus Environment

Page 3 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

both computer simulations and physical experiments.

2631

Generally, one of the most used

strategies for the model parameters identication and/or estimation involves an o-line optimization using a nominal model of the process.

3240

The main disadvantage of this method-

ology is that the variability of microorganisms decreases the possibility of batch to batch repeatability. Other techniques that are used for model parameters estimation, which try to improve the results obtained with nominal optimization methodologies, are those called "run to run" optimization, in which the information is extracted from previous runs and used to optimize the operation of subsequent ones.

4148

However, the value of this improvement should be

critically evaluated regarding to the low-variability objective that is so important in the pharmaceutical and polymer industries.

49

In the last two decades, several research projects have ventured into on-line optimization of the model parameters.

5054

This kind of optimization is dicult to perform since the

available models might only be locally valid and thus inappropriate for predicting nal concentrations.

49

On the other hand, several feedback control strategies are studied to deal with bioprocesses uncertainties.

Adaptive control techniques, for example, are an important option

to control bioreactors under uncertainties. cessfully applied to several bioreactors.

29,5557

58,59

Also, fuzzy control systems have been suc-

Alternative methodologies for controlling biopro-

cesses under model uncertainties are trajectory-based control, and hybrid control.

62

60

model predictive control

61

However, the application of this techniques in a real plant is quite

dicult because of its complex design and implementation. For the particular case of recombinant protein production with a double feed stream (Lee-Ramirez bioreactor

63

), Pantano et al

of predened optimal proles.

1

developed a control strategy for tracking control

In that work, the control design include a neural network

state estimation for unmeasurable variables, enabling the closed loop implementation. The methodology is characterized by its simplicity, versatility and accuracy. Advanced mathe-

3

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 36

matical knowledge is not required for design, only basic understanding of linear algebra is needed. The main technique advantage is that control actions compute are obtained solving a linear system equations, ensuring the convergence to zero of tracking errors. Besides, that control technique has been applied successfully in several systems,

5,6469

therefore has a high

potential for applicability in all bioprocess environments. The objective of this work is to design an improved control strategy to achieve the tracking control of optimal protein concentration, cell density and volume proles obtained in,

70

with

a minimum tracking error, when the system is subjected to parametric uncertainty and perturbations in the process. To achieve this goal, a term of uncertainty is incorporated to the original controller design, which is used to represent a wide range of model mismatches as well as perturbations in the process.

Then, some integral terms of tracking error are

incorporated into the controller design to compensate the uncertainty. In a similar way to the original methodology, the necessary and sucient conditions are analyzed so that the system has an exact solution, but now, taking into account the new terms.

Finally, the

control actions are found solving a linear system equations. The main contribution of this work is the extension of the proposed methodology in Pantano et al,

1

to provide a positive answer to the challenging problem of tracking control

in multivariable nonlinear systems with additive uncertainty and process disturbances. In this way, a more realistic problem can be solved taking account that a complete and exact knowledge of the process model is never possible and usually, the external perturbations in the real process dynamics are unavoidable. It is important to emphasize that this approach is achieved without signicantly increase of the controller design complexity. Also, another important contribution is the demonstration of convergence to zero of the tracking error under parametric uncertainties and process disturbances. The paper is organized as follows. presented in Section 2.

First, a review of the original controller design is

Then, the extended controller methodology for uncertainties con-

templation is developed in Section 3.

The results and discussion of the simulation tests,

4

ACS Paragon Plus Environment

Page 5 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

which include the adjustment of the parameters of the controller and the control system under parametric uncertainty and perturbations in the control actions are shown in Section 5. Finally, Section 6 outlines the main conclusions of this work.

Original Controller Design The controller methodology proposed in

1

is mainly based on approximating the mathematical

model (1) by employing Euler method, which, despite its simplicity, presents good results. The aim of this controller design is to nd the control actions values that follow desired paths with minimal tracking error. The case study proposed in

1

for control, is the Lee-Ramirez fed-batch bioreactor. The

mathematical model is taken of Tholudur and Ramirez.

71

The fed-batch bioreactor is described by the following model (1).

   x˙ 1 = u1 + u2      u1 +u2   x˙ 2 = x2 µ − x1 x2      µ u1 +u2 u1 N x˙ 3 = x1 − x1 x3 − Y x2     u1 +u2 x˙ 4 = x2 R − x1 x4     u2 I u1 +u2  x˙ 5 = x1 − x1 x5        x˙ 6 = −K1 x6       x˙ 7 = K2 (1 − x7 )

(1)

Where,

µ=

KCN R=

µmax x3 + x3 1 +

KCN

! x3 Ks



0 fmax x3 + x3 1 +

K1 = K2 =

5

KCI x6 + x7 KCI + x5 !

x3 Ks



fI0 + x5 K I + x5

k11 x5 KIX + x5

ACS Paragon Plus Environment

 (2)

 (3)

(4)

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

A more detailed description of the process can be found in. eters set were validated and presented by Lee and Ramirez.

(L/h),

and inducer feeding rate

for the fed-batch bioreactor. concentration

x4

u2 (L/h)

72

1

Page 6 of 36

The nominal model param-

The glucose feeding rate,

u1

are the two available sources as control actions

The reactor volume

x1 ,

cell density

x2 ,

and foreign protein

are the variables whose optimal proles is proposed to follow.

Controller Methodology A Dierential Equations System can be approximated according to numerical integration rule of Euler as follow:

x˙ i = Where

xi

is the

xi,n+1 − xi,n Ts

i state variable in n and n + 1 time instants,

(5)

being

Ts

the sampling time.

Although there exist general numerical computation algorithms in the literature for model approximation,

47,73,74

the Euler rule is a proper choice for this case since the approximation

is only used to nd the best manner to go from the current state to the following one, and not to duplicate the entire system evolution. Taking into account eq 5, the mathematical model that represents the bioprocess can be rewritten, as follows:

6

ACS Paragon Plus Environment

Page 7 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

zn+1

z

z }| { z }|n {  x x  1, n+1   1, n      x2,  x2,   n+1   n      x     3, n+1  = x3, n  +...         x4, n+1  x4, n      x5, n+1 x5, n   ξn (zn ) βn (zn ,un ) }| { z }| { z      0 1 1      u n      z }| { x x   2, n    − 2, n −   x2, n µ  x1, n    x1, n      u 1, n   x2, n µ   (N −x3,n ) x3,n   ... + − Y  +  x   Ts − x1,     n    1, n u  2, n   x4, n x4, n    x2, R  − x1,   − x1,   n  n n      x (I−x5,n )   0 − x5,1, n   x1, n

Denote by tively.

ξn (zn )

zn and

and

zn+1

(6)

n

the state vector at the current time and at the next one respec-

βn (zn , un )

are the input matrices and

un

the control actions vector.

Then, in a generic form, the system can be expressed as follows:

zn+1 = zn + [ξ(zn ) + β(zn , un )un ] Ts

(7)

Now, assuming a proportional approaching to the error:

e

en+1 }|n { z z }| { zref,n+1 − zn+1 = K (zref,n − zn )

7

ACS Paragon Plus Environment

(8)

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 36

With

zref,n+1

 zref,n  zn+1 K zn }| { }| { }| { z }| {  z z z       x1ref, n+1 x1, n+1 k1 0 0 0 0  x x 1ref, 1, n n                    x2ref,  x2,   0 k2 0 0 0   x2ref,  x2,   n+1  n   n+1    n              x  − x  =  0 0 k 0 0  x  − x  3  3ref, n+1   3, n+1     3ref, n   3, n                      x4ref, n+1  x4, n+1   0 0 0 k4 0  x4ref, n  x4, n              x5ref, n+1 x5, n+1 0 0 0 0 k5  x5ref, n x5, n   z 

Where,

}|

xi,ref,n

proles in the for the

i

{

n

and

xi,ref,n+1

(9)

are the reference values obtained from the optimal operating

time and the next sample time respectively,

variable and K is the control parameters matrix;

en

ki

and

is the controller parameter

en+1

are the tracking error

(dierence between the reference and actual proles). Then, the immediately reachable value for the state variables is:

e

zn+1

}|n { z = zref,n+1 − K (zref,n − zn )

(10)

It is important to remark that, the optimal proles to follow were obtained by an openloop simulation of the bioprocess using the optimal feeding policies achieved by Balsa Canto et al.

70

In this way, the actual state variables in the following sampling time (xi,n+1 ) only depend on the reference proles, the actual state variables at the current time and the controller parameters (all values are known). Consequently, substituting (10) in (eq 6) and rearranging the system equations:

8

ACS Paragon Plus Environment

Page 9 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

A

}|

z

{

1 1   u   z }| {   −1 −1     u1, n  (N − x )  = −x 3,n 3,n     u2, n   −1 −1     −1 (I − x5,n ) b

z



}|

x1,ref, n+1 −k1 (x1,ref, n −x1, n )−x1, n Ts

(11)

{



        x2,ref, n+1 −k2 (x2,ref ,n −x2, n )−x2, n x 1, n  − x2, n µ  Ts x2, n       x3,ez, n+1 −k3 (x3,ez, n −x3, n )−x3, n  µ  x + x x = 1, 2, 1, n n n  Ts Y      x4,ref, n+1 −k4 (x4,ref, n −x4, n )−x4, n x1,  x 1, n n  − Rx 2, n x4, n  Ts x4, n        x5,ez, n+1 −k5 (x5,ez, n −x5, n )−x5, n x 1, n Ts The optimal proles to follow are those corresponding to The only unknown variables of this system are dened as  sponding in this case to

x3,ez

and

x5,ez .

x1 , x 2

and

x4 ,

that are known.

sacriced variables  xi,ez ,

corre-

As can be seen in eq 11, the system normally has

no solution (ve equations and two unknowns). Therefore, the key of the control technique proposed in

1

is that the values adopted by 

sacriced variables 

forces the equation system

(11) to have exact solution, which implies the tracking error not only to be minimal, but equal zero. As mentioned above, the equations system (11) has not exact solution, therefore to accomplish the target of this control methodology, that system must have exact solution. Then, a Gaussian elimination process is carried out to nd the necessary and sucient condition for the system to have an exact solution (the resultant expression can be seen

1

in ). Therefore, the unknown variables,

x3,ez

and

x5,ez (sacriced variables )

are computed

for each sampling time. Once the values of sacriced variables are found, the system (11) has an exact solution and can be solved by the least squares method for the calculation of

9

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the control variables (u1,n and

u2,n , u

Page 10 of 36

vector).

−1

(AT A)u = AT b ⇒ u = (AT A) AT b The control actions (u1,n and

(12)

u2,n ) applied at time n allow to follow the desired trajectories

with a minimal error.

Tracking Error The tracking error is dened as the value of the dierence between the reference and real trajectory. In a generic way, the tracking error for each state variable is dened as:

ei,n = xi,ref,n − xi,n

,for i=1,2,3,4,5

(13)

The dimensionless tracking error (for desired variables) at n instant of time is calculated as:

s kead,n k =

2

e1,n x1,ref,max

 +

e2,n

2

x2,ref,max

 +

e4,n

2 (14)

x4,ref,max

and, the Total Tracking Error

ET =Ts

X

kead,n k

(15)

n

The reference nal values for the desired variables are:

13, 92g/L,

and

x1,ref,max = 1, 9L, x2,ref,max =

x4,ref,max = 3, 1g/L.

Note that the total tracking error (TTE, eq 15) is dimensionless. Taking into account eq 8 and eq 11, the original control system proposed in

1

(that did

not take into account parametric uncertainties or process perturbations for the controller design) can be denoted as:

10

ACS Paragon Plus Environment

Page 11 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

en+1

e

z }| {  z }|n {  e k 0 0 0 0 e  1,n+1   1   1,n       e2,n+1   0 k2 0 0 0  e2,n            e      3,n+1  =  0 0 k3 0 0  e3,n  +...           e4,n+1   0 0 0 k4 0  e4,n       e5,n+1 0 0 0 0 k5 e5,n

(16)

BOUNDED NONLINEARITY → ZERO

}|

z

{

0 0      −x2 e3,ez,n µλ −x2 e5,ez,n µθ       ... + Ts  0 0       −x2 e3,ez,n Rλ −x2 e5,ez,n Rθ    0 0

Equation 16 demonstrate that the tracking error for all variables tend to zero when

0 < ki < 1, i = 1, 2, 3, 4, 5,

and

n → ∞.

Demonstration can be seen in.

1

Controller design under parametric uncertainty and process disturbances In this section, a methodology for the parametric uncertainties and process disturbances handling is presented. In order to quantify the model mismatch and process disturbances, an additive uncertainty is incorporated to the original controller design. According to eq 7, the real bioprocess model is assumed:

Real System

z }| { Model z }| { zn+1 = zn + [ξ(zn ) + β(zn , un )un ] T s + En Where,

En

(17)

quantify the uncertainty. Note that this term of uncertainty can be used to

11

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 36

model a wide class of model mismatches as well as perturbed systems. The mismatch model and external perturbations might depend on the state variables or the system input.

Therefore, considering a real plant:

uncertainty can be expressed as

En = g(zn , un ) − gˆ(zn , un ),

system model in discrete time. Now, note that if it will be assumed, then

En

z

and

u

zn+1 = g(zn , un ), where

gˆ(zn , un )

are bounded and

can be modeled as a bounded uncertainty.

the additive

is the nonlinear

g

is Lipschitz, as

75,76

Analogously to the procedure followed for eq 16, but now taking into account the additive uncertainty, it is obtained the next concluding expression:



e1,n+1





k1

      e2,n+1   0       e    3,n+1  =  0       e4,n+1   0    e5,n+1 0

0

0

0

k2

0

0

0

k3

0

0

0

k4

0

0

0

0



E

z }|n {  0 E   1,n       −x2 e5,ez,n gθ   E2,n      − E  0   3,n        −x2 e5,ez,n Rθ  E4,n     0 E5,n

}|

0    −x2 e3,ez,n gλ   + ...Ts  0    −x2 e3,ez,n Rλ  0



      0  e2,n      0  e3,n  + ...     0  e4,n    k5 e5,n

BOUNDED NONLINEARITY → ZERO

z 

e1,n

(18)

{

Looking at eq 18, it can be seen how the additive uncertainty directly aects the tracking error. Anyway, the presence of the term

En

makes the tracking error not to converge to zero.

The main contribution of this work is to compensate the uncertainty and achieve the tracking error convergence to zero when the process moves forward. Therefore, the objective is to compensate the uncertainty and achieve the tracking error convergence to zero when the process moves forward.

12

ACS Paragon Plus Environment

Page 13 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Integral action In order to deal with the additive uncertainty eect on the tracking error, the incorporation of an integral action in the state variables system is proposed. Therefore, depending on the supposition of the time variation of

En ,

a series of the tracking errors integrators are added

in the control actions calculation. In a real system, it is assumed that the eects of additive uncertainties on tracking errors are unknown and each component of

Eϕ,n

can be represented by a polynomial of

As denition, the rst order dierence for additive uncertainty is second order dierence

δ q En = δ (δ q−1 En ). 77

order.

δEn = En+1 − En ,

δ 2 En = δ (δEn ) = δ (En+1 − En ) = En+2 − 2En+1 + En ,

the q-th order dierence can be expressed as

m

the

analogous,

Note that, the q-th order

dierence of a q-1 polynomial order is zero.

Constant uncertainty. If uncertainties remains constant throughout the process, ie dierence is equal to zero:

En = const,

then the rst order

δEn = En+1 − En = 0

The integral action proposed is dened as:

(n+1)T Z S

U1,n+1 = U1,n +

e(t)dt ∼ = U1,n + en TS

(19)

nTS

Where,

e(t)

is the continuous time error in the state vector and

the error. The subscript

1

means the st integral of error, and

n

U1,n+1

is the integral of

the instant of time.

According to eq 10, but now considering parametric uncertainties and process disturbances, the control action can be computed taking into account the new terms:

e

zn+1

Where

L1

}|n { z = zref,n+1 − K (zref,n − zn ) +L1 U1,n+1

(20)

is the matrix corresponding to integral action with a constant uncertainty.

13

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 36

Then, for each variable:

 zref,n  zn+1 K zn }| { }| { }| { z }| { z z z        x x k 0 0 0 0  x1ref, n x  1ref, n+1   1, n+1   1     1, n             x2ref,  x2,   0 k2 0 0 0      x2ref, n  x2, n   n+1    n+1                 x      x    3ref, n  − x3, n  + ...  3ref, n+1  − x3, n+1  =  0 0 k3 0 0                        x4ref, n+1  x4, n+1   0 0 0 k4 0   x4ref, n  x4, n               x5ref, n+1 x5, n+1 0 0 0 0 k5  x x  5ref, n 5, n  zref,n+1

z

}|

{

U1,n

L

}|1 z { z }| { L 0 0 0 0 U  1,1   1,1      0 L1,2 0  U1,2  0 0         ... +  0 L1,3 0 0   0  U1,3        0 0 L1,4 0  U1,4   0    0 0 0 0 L1,5 U1,5 | {z }

(21)

CONSTANT INTEGRAL ACTION

Now, following the procedure carried out for the original controller, eq 21 is replaced in the mathematical model of the system represented in eq 6:

14

ACS Paragon Plus Environment

Page 15 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research







1 1       −1 −1      u1, n  (N − x )  = ... −x3,n  3,n     u2, n   −1 −1     −1 (I − x5,n )   x1,ref, n+1 −k1 (x1,ref, n −x1, n )+L1,1 U1,1,n+1 −x1, n



Ts         x2,ref, n+1 −k2 (x2,ref ,n −x2, n )+L1,2 U1,2,n+1 −x2, n x1, n  − x2, n µ  Ts x2, n        x3,ez, n+1 −k3 (x3,ez, n −x3, n )+L1,3 U1,3,n+1 −x3, n µ x1, n + Y x2, n x1, n  ... =  Ts       x4,ref, n+1 −k4 (x4,ref, n −x4, n )+L1,4 U1,4,n+1 −x4, n x1,  x 1, n n  − Rx 2, n Ts x4, n x4, n         x5,ez, n+1 −k5 (x5,ez, n −x5, n )+L1,5 U1,5,n+1 −x5, n x 1, n Ts

(22)

The next step is to nd the necessary condition for the system (eq 22) to have an exact solution, which is achieved through a Gaussian elimination process similar to that presented in.

1

In this way, the values of the variables sacriced for each sampling time are found. Then,

the control actions can be calculated using eq 12. Now, to demonstrate that the tracking error of the real system under perturbations converges to zero, a simple mathematical operations are carried out: Expressing eq. 18 in the following summarized way:

en+1 = −Ken + N Ln − E1,n+1

(23)

en+1 = −Ken + N Ln + L1 U1,n+1 − E1,n+1

(24)

Adding the integral term,

15

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 36

Replacing eq 19 in eq 24,

en+1 = Ken + N Ln + L1 U1,n + L1 en Ts + E1,n+1

(25)

Increasing eq 25 in one sampling time:

 en+2 = Ken+1 + N Ln+1 + L1 U 1,n+1 + en+1 Ts + E1,n+2

(26)

Clearing the integral term from eq 24, replacing in 26 and rearranging:

δE2,n =0

en+2

z }| { = −en+1 (−1 − K + L1 Ts ) − Ken − (N Ln+1 − N Ln ) + (E2,n+1 − E2,n )

Therefore, the set of parameters eq 27. The Non Linearity term

K

and

L1

(27)

are optimized for the stability assurance of

(N Ln+1 − N Ln )

tends to zero because depends on the error

in the sacriced variables, which tends to zero too.

Then

en → 0

as

n → ∞

despite the

uncertainties, if they are considered constant.

Linear uncertainty If uncertainties can be represented by a linear function, where the second order dierence is equal to zero:

δ 2 En = δ (δEn ) = δ (En+1 − En ) = En+2 − 2En+1 + En = 0,

then a new

integrator must be considered. In the similar way that procedure before, but now introducing two integral actions dened by

U1

and

U2 :

(n+1)T Z S

Uϕ,2,n+1 = Uϕ,2,n +

Uϕ,1 (t)dt ∼ = Uϕ,2,n + Uϕ,1,n+1 TS

nTS Where, the subscript

2

means an integrator for a linear perturbation.

Then, a new term is added to eq 20:

16

ACS Paragon Plus Environment

(28)

Page 17 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

e

zn+1 The new parameter

z }|n { = zref,n+1 − K (zref,n − zn ) +L1 U1,n+1 + L2 U2,n+1 L2

(29)

correspond to double integral action. In the similar way as simple

integral action, the demonstration of convergence to zero of tracking error for the double integral action is: Operating as before, and taking into account that

δ 2 En = 0 ,

the nal

expression for tracking error is:

en+3 = −en+2 (−K + Ts (L1 + L2 Ts ) − 2) − en+1 (2K − L1 Ts + 1) + ... δ 2 E2,n =0

BOUNDED NONLINEARITY

(30)

}| { z }| { z ... + Ken (N Ln+2 − N Ln+1 + N Ln ) + (En+2 − 2En+1 + E2,n ) Now, as can be seen in (30), under constant or linear varying uncertainty, this has no inuence on the error dynamics. The parameters

K, L1 , L2

are optimized for the stability

assurance of eq 30, as shown in the previous case. Now, in a generic way, if the uncertainties can be approximated by the inuence of

En

on

en

will be eliminated by introducing

q

q−1 order polynomial,

integrators.

Optimization of controller parameters There is a recurrent problem for this kind of controllers and it is how to dene the best tuning parameters to achieve a good closed-loop response. In this subsection an algorithm based on Montecarlo randomized experiment is proposed to nd the optimal controller parameters. In the original controller presented in,

K = {k1 , k2 , k3 , k4 , k5 } , eq (27) are that

1

the controller parameters are represented by

and the conditions for the tracking error tends to zero according to

0 < ki < 1, i = {1, 2, 3, 4, 5 } .

But now, integral actions are added, therefore

the amount of parameters to select increases and the conditions change:

K(Original Controller) L1 (Integral action, constant uncertainty) L2 (Integral action, linear uncertainty)

z }| { k1 , k2 , k3 , k4 , k5 ,

z }| { L1,1 , L1,2 , L1,3 , L1,4 , L1,5

17

,

ACS Paragon Plus Environment

z }| { L2,1 , L2,2 , L2,3 , L2,4 , L2,5

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

In a generic form, for the integral action, the parameters are denoted by is the number of integrators and

i

Page 18 of 36

Lm,i ,

where

m

correspond to each state variable.

To choose the optimization criteria of the parameters considering the integral actions, it is resort to the characteristic equations (27) and (30) (simple and double integral action respectively). In the case of the controller without integral action, the characteristic equation can be rewritten as:

r − ki = 0

(31)

Therefore, the parameters are directly the roots of polynomial. For a simple integral action (one integrator, constant uncertainty), the characteristic equation is (27) and can be rewritten as:

r2 + r (−1 − ki + L1,i Ts ) + ki = 0

(32)

Thus clearing the parameters according to the roots:

ki = r1 r2 (33)

L1,i = (−r1 − r2 + r1 r2 + 1)/Ts For a double integral action (two integrators, linear uncertainty), the characteristic equation is (30) and can be rewritten as:

r3 + r2 (−ki + Ts (L1,i + L2,i Ts ) − 2) + r (2ki − L1,i Ts + 1) − ki = 0

Clearing parameters:

18

ACS Paragon Plus Environment

(34)

Page 19 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

ki = r1 r2 r3 (35)

L1,i = (2r1 r2 r3 + 1 − r1 r2 − r1 r3 − r2 r3 )/Ts L2,i = (−r1 − r2 − r3 − r1 r2 r3 + 1 + r1 r2 + r1 r3 + r2 r3 )/Ts2

Thus, for the tracking error tends to zero and ensure the system stability, the roots of the polynomial must be between 0 and 1. The performance index used to optimize the controller parameters is to minimize the total tracking error (15):

C = min (ET )

(36)

K,L

Results and Discussion In this section, the behavior of the controller against parametric uncertainty and perturbations in the process is evaluated through simulations. First, a Montecarlo algorithm to tune the optimal controller parameters is carried out. Then, the system is tested under parametric uncertainty and process disturbances, comparing the system response with and without integral action.

Controller tuning In order to determinate the optimal controller parameters, a Montecarlo algorithm (MCRA) is carried out.

This methodology has been recently employed for controllers tuning

1,64,78

with satisfactory results. According to the criteria for parameters selection discussed in previous section, the procedure for the simulation is detailed below:

Controller tuning without integral action (m = 0): is the same as in,

1

in this case, the methodology employed

where 1000 simulations are performed.

19

ACS Paragon Plus Environment

Considering eq 31, for each

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

simulation a random value

ri  (0, 1)

Page 20 of 36

is assigned for each state variable, then the control

actions and the performance index (36) are computed. Once the simulations are nished, the optimal controller parameters set is the corresponding to the minimum

Controller tuning with an integral action (m = 1):

C.

again, 1000 simulations are performed,

but now the parameters are calculate according eq 33.

Controller tuning with a double integral action (m = 2):

once the 1000 simulations are

performed, the controller parameters are compute according eq 35. The necessary information to carry out the simulations is taken from. sampling time are

Tf = 10h

and

T s = 0.1h

1

The nal and

respectively.

Table 1 shows the optimized values for the controller parameters. Depending on the value of

m

there will be ve, ten or fteen parameters. Table 1: Optimal controller parameters Parameters

m=0

m=1

m=2

k1 k2 k3 k4 k5 L1,1 L1,2 L1,3 L1,4 L1,5 L2,1 L2,2 L2,3 L2,4 L2,5

0.2805

0.0038

0.0404

0.6761

0.9412

0.9412

0.6520

0.7941

0.6589

0.1694

0.2032

0.3842

0.1266

0.5382

0.5382

-

8.8102

0.0851

-

0.0088

2.3546

-

1.1183

1.2538

-

1.4755

2.4587

-

0.6504

1.0085

-

-

0.2588

-

-

1.5285

-

-

3.0497

-

-

1.9246

-

-

2.7533

Disturbances and uncertainties handling. For the uncertainties and disturbances quantication in the simulation work, an approach consists of specifying only the upper and lower limits within which the real disturbance or

20

ACS Paragon Plus Environment

Page 21 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

uncertainty is assumed to evolve. The idea of this approach is thus to replace the knowledge of exact values of unknown inputs by a known range of the values within which they evolve.

79

Parametric uncertainty In the bioprocesses, the model parameters vary in an unpredictable manner. to a structural instability in the dynamical behavior of the system.

20

80

It can lead

Therefore, strict and

ecient control system is necessary. In several research elds, probabilistic methods have been found to be useful for dealing with problems related to robustness of systems aected by uncertainties.

81

In particular,

the Monte Carlo Randomized Algorithm has been used for uncertainty quantication in many applications such as Rothermel model,

48

power system generation and trac on roads,

river ow rate forecast,

83

82

radioactive decay,

among others. From the point of view of

process control, Monte Carlo methods are eective tools for the analysis of probabilistically robust control schemes.

65,81

In this subsection the system is simulated considering errors on

modeling using the MCRA Method.

The procedure consist on replacing the exact model

parameters values by a range of them within which they can vary. The dened range for model parameters is

± 20% of they nominal values.

Then, N=1000 simulations are executed

using the optimized controller parameters set. Taking account the worst case of uncertainty, all the model parameters are varied. In the rst instance, the control system is tested without integral action.

Then one

integrator is added and nally, two integrators are added. The results are compared below. The tracking of optimal proles for the three cases is shown in g 1. Figure 2 shows the TTE for 1000 simulations to evaluate the controller performance under parametric uncertainty for three cases: none integrator (a), one integrator (b) and two integrators (c). As can be seen, the total tracking error range is visibly reduced when integral actions are added. In Table 2, the total tracking error for the three cases is shown. Now, taking the average

21

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 36

Figure 1: Tracking of optimal proles for 1000 simulations under parametric uncertainty

(±20%).

(a) Non integral action

(m = 0),

(b) One integral action

(m = 1)

and (c) Two

integral actions. Dashed line shows the reference values.

(±20%). (m = 0), (b) One integral action (m = 1)

Figure 2: Total Tracking Error for 1000 simulations under parametric uncertainty (a) Non integral action, proposed in Pantano et al. and (c) Two integral actions

(m = 2).

1

Central dashed line shows the mean values.

22

ACS Paragon Plus Environment

Page 23 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

value of TTE for each case and traducing it to percentage, it can be easily quantied the improvement of the controller with integral action (see g 3). Table 2: Reduction of TTE by adding integrators under parametric uncertainty m=0

m=1

m=2

TTE

0.0472

0.0245

0.0186

%

100

52

39

Figure 3: TTE mean value for the three cases. Process simulation under parametric uncertainty (20%).

Note that the error is reduced by

48%

by adding one integral action and

61%

with two

integrators. The above gures show that the integral actions incorporated into the controller are eective and give robustness to the control system.

Perturbation in the control actions In all processes there are also disturbances acting on the plant in addition to parametric uncertainties, modifying the expected values of the system variables. Therefore, a random perturbation in the control actions taking into account parametric uncertainties is simulated for the controller evaluation. The two control actions are perturbed by values:

u1,perturbed = u1,unperturbed (1 + random(unif, 0, 0.3) + 1) 23

ACS Paragon Plus Environment

30 % of its calculated

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 36

u2,perturbed = u2,unperturbed (1 + random(unif, 0, 0.3) + 1) Figure 4 shows the controller adjustment to hold the tracking error in a minimal value spite of perturbations in the control actions a parametric uncertainties.

Figure 4: Perturbed control actions considering parametric uncertainty.

Table 3 shows the TTE for the controllers. Note how again, by adding only one integral action the tracking error is reduced by a

68 %.

The response improves by a

75 %

with two

integrators. Table 3: TTE for perturbed system with parametric uncertainty. m=0

m=1

m=2

TTE

0.1036

0.0335

0.0253

%

100

32

24

Figure 5: TTE for the three cases.

Process simulation under parametric uncertainty and

perturbations in the control actions.

Figure 5 shows a bar diagram of the cumulative error (TTE) for the real system under parametric uncertainties and perturbation in the control actions. This is presented for the

24

ACS Paragon Plus Environment

Page 25 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

original controller proposed in Pantano et al, in this paper:

1

(m = 0), and for the two controllers proposed

one integrator (m = 1) and two integrators (m = 2).

In addition, as a

comparison, TTE is shown for a classic PI controller, which was automatically tuned by a Matlab auto-tuning function. As can be seen in this section, the simulation work demonstrate the eectiveness of the controller proposed in this work, by adding integral actions to the design, in order to solve the problems of control under parametric uncertainty and process disturbances.

Conclusion This paper proposes an improved control system for fed-batch production of recombinant protein, a multivariable and nonlinear bioprocess which is highly susceptible to model and process disturbances due to its biological nature. The original controller proposed in

1

achieved

the tracking of optimal proles of volume, cell density and protein concentration (main product) with a minimum error through the development of an eective control law based on linear algebra. The technique proposed in this work extends the original controller design to take into account parametric uncertainty and process disturbances. Several simulations were carried out to test the controller performance. The system was evaluated under parametric uncertainties

(±20%)

and random perturbations in the control actions

the system response up to

70 %

(±30%),

improving

by adding some integral actions of the tracking error in the

control actions compute. The optimal controller parameters (with, and without integrators) were successfully found through a Montecarlo experiment. Although this method has low complexity, the results are reliable and solves a real problem for bioprocess control.

Author Information ∗

E-mail: [email protected]

25

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 26 of 36

Acknowledgment This work was partially funded by the following Argentine institutions: National Council of Scientic and Technological Research (CONICET) and Universidad Nacional de San Juan (UNSJ).

References (1) Pantano, M. N.; Serrano, M. E.; Fernández, M. C.; Rossomando, F. G.; Ortiz, O. A.; Scaglia, G. J. Multivariable Control for Tracking Optimal Proles in a Nonlinear FedBatch Bioprocess Integrated with State Estimation.

Industrial & Engineering Chem-

istry Research 2017, 56, 60436056. (2) Simutis, R.; Lübbert, A. Bioreactor control improves bioprocess performance.

Biotech-

nology journal 2015, 10, 11151130. (3) Dimitrova, N.; Krastanov, M. Nonlinear adaptive control of a model of an uncertain fermentation process.

International Journal of Robust and Nonlinear Control 2010, 20,

10011009.

(4) Beltran, L. M.; Garzon-Castro, C. L.; Garces, F.; Moreno, M. Monitoring and control system used in microalgae crop.

IEEE Latin America Transactions 2012, 10,

1993

1998.

(5) Rómoli, S.; Amicarelli, A. N.; Ortiz, O. A.; Scaglia, G. J. E.; di Sciascio, F. Nonlinear control of the dissolved oxygen concentration integrated with a biomass estimator for production of Bacillus thuringiensis

δ -endotoxins. Computers & Chemical Engineering

2016, 93, 1324. (6) Carcamo, M.; Saa, P.; Torres, J.; Torres, S.; Mandujano, P.; Correa, J. R. P.; Agosin, E.

26

ACS Paragon Plus Environment

Page 27 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

Eective dissolved oxygen control strategy for high-cell-density cultures.

IEEE Latin

America Transactions 2014, 12, 389394. (7) Veluvolu, K.; Soh, Y.; Cao, W. Robust discrete-time nonlinear sliding mode state estimation of uncertain nonlinear systems.

International Journal of Robust and Nonlinear

Control 2007, 17, 803828. (8) Rathore, A. S.; Winkle, H. Quality by design for biopharmaceuticals.

Nature biotech-

nology 2009, 27, 26. (9) Gnoth, S.; Jenzsch, M.; Simutis, R.; Lübbert, A. Control of cultivation processes for recombinant protein production: a review.

Bioprocess and biosystems engineering 2008,

31, 2139. (10) Imtiaz, U.; Jamuar, S. S.; Sahu, J.; Ganesan, P. Bioreactor prole control by a nonlinear auto regressive moving average neuro and two degree of freedom PID controllers.

Journal of Process Control 2014, 24, 17611777. (11) Kuprijanov, A.; Schaepe, S.; Aehle, M.; Simutis, R.; Lübbert, A. Improving cultivation processes for recombinant protein production.

Bioprocess and biosystems engineering

2012, 35, 333340. (12) Valentinotti, S.; Srinivasan, B.; Holmberg, U.; Bonvin, D.; Cannizzaro, C.; Rhiel, M.; Von Stockar, U. Optimal operation of fed-batch fermentations via adaptive control of overow metabolite.

Control engineering practice 2003, 11, 665674.

(13) De Battista, H.; Picó, J.; Picó-Marco, E. Nonlinear PI control of fed-batch processes for growth rate regulation.

Journal of Process Control 2012, 22, 789797.

(14) Jin, H.; Chen, X.; Yang, J.; Wu, L.; Wang, L. Hybrid intelligent control of substrate feeding for industrial fed-batch chlortetracycline fermentation process.

2014, 53, 18221837. 27

ACS Paragon Plus Environment

ISA transactions

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 28 of 36

(15) Terwiesch, P.; Agarwal, M.; Rippin, D. W. Batch unit optimization with imperfect modelling: a survey.

Journal of Process Control 1994, 4, 238258.

(16) Terwiesch, P.; Ravemark, D.; Schenker, B.; Rippin, D. W. Semi-batch process optimization under uncertainty: Theory and experiments.

Computers & Chemical Engineering

1998, 22, 201213. (17) Srinivasan, B.; Biegler, L. T.; Bonvin, D. Tracking the necessary conditions of optimality with changing set of active constraints using a barrier-penalty function.

Computers

& Chemical Engineering 2008, 32, 572579. (18) Nagy, Z. K.; Braatz, R. D. Open-loop and closed-loop robust optimal control of batch processes using distributional and worst-case analysis.

Journal of process control 2004,

14, 411422. (19) Hess, J.; Bernard, O. Design and study of a risk management criterion for an unstable anaerobic wastewater treatment process.

Journal of Process Control 2008, 18, 7179.

(20) Lara-Cisneros, G.; Femat, R.; Pérez, E. On dynamical behaviour of two-dimensional biological reactors.

International Journal of Systems Science 2012, 43, 526534.

(21) Yao, J.; Jiao, Z.; Ma, D.; Yan, L. High-accuracy tracking control of hydraulic rotary actuators with modeling uncertainties.

IEEE/ASME Transactions on Mechatronics 2014,

19, 633641. (22) Apkarian, P.; Dao, M. N.; Noll, D. Parametric robust structured control design.

IEEE

Transactions on automatic Control 2015, 60, 18571869. (23) Alvarez-Ramírez, J. Adaptive control of feedback linearizable systems: error compensation approach.

a modelling

International Journal of Robust and Nonlinear Control

1999, 9, 361377.

28

ACS Paragon Plus Environment

Page 29 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

(24) Méndez-Acosta, H. O.; Palacios-Ruiz, B.; Alcaraz-Gonzalez, V.; Steyer, J.-P.; GonzalezAlvarez, V.; Latrille, E. Robust control of volatile fatty acids in anaerobic digestion processes.

Industrial & Engineering Chemistry Research 2008, 47, 77157720.

(25) Allen, A. P.; Gillooly, J. F. Assessing latitudinal gradients in speciation rates and biodiversity at the global scale.

Ecology letters 2006, 9, 947954.

(26) Riley, M. E.; Grandhi, R. V. Quantication of model-form and predictive uncertainty for multi-physics simulation.

Computers & Structures 2011, 89, 12061213.

(27) Xiong, L.; Wan, M.; Wei, X.; O'connor, K. M. Indices for assessing the prediction bounds of hydrological models and application by generalised likelihood uncertainty estimation/Indices pour évaluer les bornes de prévision de modèles hydrologiques et mise en ÷uvre pour une estimation d'incertitude par vraisemblance généralisée.

Hydrological

Sciences Journal 2009, 54, 852871. (28) Park, S. H.; Blackstone, C. Further assembly required: construction and dynamics of the endoplasmic reticulum network.

EMBO reports 2010, 11, 515521.

(29) Keesman, K. J.; Maksimov, V. On feedback identication of unknown characteristics: a bioreactor case study.

International Journal of Control 2008, 81, 134145.

(30) Mandur, J.; Budman, H. Robust optimization of chemical processes using Bayesian description of parametric uncertainty.

Journal of Process Control 2014, 24, 422430.

(31) Boon, M.; Janssen, A.; Van der Padt, A. Modelling and parameter estimation of the enzymatic synthesis of oligosaccharides by

β -galactosidase

from Bacillus circulans.

Biotechnology and bioengineering 1999, 64, 558567. (32) Donoso-Bravo,

A.;

Mailier,

J.;

Martin,

C.;

Rodríguez,

J.;

Aceves-Lara,

C.

A.;

Wouwer, A. V. Model selection, identication and validation in anaerobic digestion: a review.

Water research 2011, 45, 53475364. 29

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 30 of 36

(33) Morales-Rodriguez, R.; Meyer, A. S.; Gernaey, K. V.; Sin, G. A framework for modelbased optimization of bioprocesses under uncertainty: Lignocellulosic ethanol production case.

Computers & Chemical Engineering 2012, 42, 115129.

(34) Wechselberger, P.; Seifert, A.; Herwig, C. PAT method to gather bioprocess parameters in real-time using simple input variables and rst principle relationships.

Chemical

Engineering Science 2010, 65, 57345746. (35) Balsa-Canto, E.; Alonso, A. A.; Banga, J. R. An iterative identication procedure for dynamic modeling of biochemical networks.

BMC systems biology 2010, 4, 11.

(36) Vilas, C.; Arias-Méndez, A.; García, M. R.; Alonso, A. A.; Balsa-Canto, E. Toward predictive food process models: A protocol for parameter estimation.

Critical reviews

in food science and nutrition 2018, 58, 436449. (37) Liu, C.; Gong, Z.

Optimal control of switched systems arising in fermentation processes ;

Springer, 2016.

(38) Liu, C.; Gong, Z.; Teo, K. L.; Feng, E. Multi-objective optimization of nonlinear switched time-delay systems in fed-batch process.

Applied Mathematical Modelling

2016, 40, 1053310548. (39) Liu, C.; Gong, Z.; Teo, K. L.; Loxton, R.; Feng, E. Bi-objective dynamic optimization of a nonlinear time-delay system in microbial batch process.

Optimization Letters 2016,

116.

(40) Liu, C.; Gong, Z.; Teo, K. L.; Sun, J.; Caccetta, L. Robust multi-objective optimal switching control arising in 1, 3-propanediol microbial fed-batch process.

Nonlinear

Analysis: Hybrid Systems 2017, 25, 120. (41) Hille, R.; Mandur, J.; Budman, H. M. Robust batch-to-batch optimization in the pres-

30

ACS Paragon Plus Environment

Page 31 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

ence of model-plant mismatch and input uncertainty.

AIChE Journal 2017, 63, 2660

2670.

(42) Hille, R.; Budman, H. M. Run-to-Run Optimization of Batch Processes Using Set-Based Constraints.

IFAC-PapersOnLine 2017, 50, 46784683.

(43) Camacho, J.; Lauri, D.; Lennox, B.; Escabias, M.; Valderrama, M. Evaluation of smoothing techniques in the run to run optimization of fed-batch processes with uPLS.

Journal of Chemometrics 2015, 29, 338348.

(44) Luna, M.; Martínez, E. A Bayesian approach to run-to-run optimization of animal cell bioreactors using probabilistic tendency models.

Industrial & Engineering Chemistry

Research 2014, 53, 1725217266. (45) Ko, C.-L.; Wang, F.-S. Run-to-run fed-batch optimization for protein production using recombinant Escherichia coli.

Biochemical engineering journal 2006, 30, 279285.

(46) Hunag, W.-H.; Shieh, G. S.; Wang, F.-S. Optimization of fed-batch fermentation using mixture of sugars to produce ethanol.

Journal of the Taiwan Institute of Chemical

Engineers 2012, 43, 18. (47) Lin, Q.; Loxton, R.; Xu, C.; Teo, K. L. Parameter estimation for nonlinear time-delay systems with noisy output measurements.

Automatica 2015, 60, 4856.

(48) Liu, Y.; Jimenez, E.; Hussaini, M. Y.; Ökten, G.; Goodrick, S. Parametric uncertainty quantication in the Rothermel model with randomised quasi-Monte Carlo methods.

International Journal of Wildland Fire 2015, 24, 307316. (49) Bonvin, D. Optimal operation of batch reactorsa personal view.

Journal of process

control 1998, 8, 355368. (50) Câmara, M.; Soares, R.; Feital, T.; Naomi, P.; Oki, S.; Thevelein, J.; Amaral, M.;

31

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 32 of 36

Pinto, J. On-line identication of fermentation processes for ethanol production.

Bio-

process and biosystems engineering 2017, 40, 9891006. (51) Goudar, C. T.; Konstantinov, K. B.; Piret, J. M. Robust parameter estimation during logistic modeling of batch and fed-batch culture kinetics.

Biotechnology progress 2009,

25, 801806. (52) Van Daele, T.; Gernaey, K. V.; Ringborg, R. H.; Börner, T.; Heintz, S.; Van Hauwermeiren, D.; Grey, C.; Krühne, U.; Adlercreutz, P.; Nopens, I. Application of iterative robust model-based optimal experimental design for the calibration of biocatalytic models.

Biotechnology progress 2017,

(53) Encarnacion-Gomez, L. G.; Bommarius, A. S.; Rousseau, R. W. Crystallization kinetics of ampicillin using online monitoring tools and robust parameter estimation.

Industrial

& Engineering Chemistry Research 2016, 55, 21532162. (54) Chae Hee Jeong, e. a. Framework for online optimization of recombinant protein expression in high-cell-density Escherichia coli cultures using GFP-fusion monitoring.

Biotech-

nology and bioengineering 2000, 69, 275285. (55) Rincón, A.; Piarpuzán, D.; Angulo, F. A new adaptive controller for bio-reactors with unknown kinetics and biomass concentration: convergence properties.

Guarantees for the boundedness and

Mathematics and Computers in Simulation 2015, 112, 113.

(56) Rodrigues, D.; Billeter, J.; Bonvin, D.

Computer Aided Chemical Engineering ; Elsevier,

2017; Vol. 40; pp 21192124.

(57) Mailleret, L.; Bernard, O.; Steyer, J.-P. Nonlinear adaptive control for bioreactors with unknown kinetics.

Automatica 2004, 40, 13791385.

(58) Tong, S.; Liu, C.; Li, Y. Fuzzy-adaptive decentralized output-feedback control for large-

32

ACS Paragon Plus Environment

Page 33 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

scale nonlinear systems with dynamical uncertainties.

IEEE Transactions on Fuzzy

Systems 2010, 18, 845861. (59) Bartolomeo, C.; Mosé, G. Type-2 fuzzy control of a bioreactor. Intelligent Computing and Intelligent Systems, 2009. ICIS 2009. IEEE International Conference on. 2009; pp 700704.

(60) Cimander, C.; Mandenius, C.-F. Bioprocess control from a multivariate process trajectory.

Bioprocess and biosystems engineering 2004, 26, 401411.

(61) Liping, F.; Jun, Z.; Xing, H.; Dong, H. The design of the MISO model predictive controller for bioreactor.

Indonesian Journal of Electrical Engineering and Computer

Science 2012, 10, 11631170. (62) Mairet, F.; Gouzé, J.-L. Hybrid control of a bioreactor with quantized measurements.

IEEE Transactions on Automatic Control 2016, 61, 13851390. (63) Lee, J.; Ramirez, W. F. Mathematical modeling of induced foreign protein production by recombinant bacteria.

Biotechnology and bioengineering 1992, 39, 635646.

(64) Scaglia, G.; Rosales, A.; Quintero, L.; Mut, V.; Agarwal, R. A linear-interpolationbased controller design for trajectory tracking of mobile robots.

Control Engineering

Practice 2010, 18, 318329. (65) Rómoli, S.; Serrano, M. E.; Ortiz, O. A.; Vega, J. R.; Scaglia, G. J. E. Tracking control of concentration proles in a fed-batch bioreactor using a linear algebra methodology.

ISA transactions 2015, 57, 162171. (66) Serrano, M. E.; Scaglia, G. J.; Godoy, S. A.; Mut, V.; Ortiz, O. A. Trajectory tracking of underactuated surface vessels:

A linear algebra approach.

Control Systems Technology 2014, 22, 11031111.

33

ACS Paragon Plus Environment

IEEE Transactions on

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 34 of 36

(67) Romoli, S.; Scaglia, G. J. E.; Serrano, M. E.; Godoy, S. A.; Ortiz, O. A.; Vega, J. R. Control of a fed-batch fermenter based on a linear algebra strategy.

IEEE Latin America

Transactions 2014, 12, 12061213. (68) Fernández, M. C.; Rómoli, S.; Pantano, M. N.; Ortiz, O. A.; Patiño, D.; Scaglia, G. J. A New Approach for Nonlinear Multivariable Fed-Batch Bioprocess Trajectory Tracking Control.

Automatic Control and Computer Sciences 2018, 52, 1324.

(69) Pantano, M. N.; Fernández, M. C.; Serrano, M. E.; Ortíz, O. A.; Scaglia, G. J. E. Trajectory Tracking Controller for a Nonlinear Fed-batch Bioprocess.

Revista Ingeniería

Electrónica, Automática y Comunicaciones ISSN: 1815-5928 2017, 38, 78. (70) Balsa-Canto, E.; Banga, J. R.; Alonso, A. A.; Vassiliadis, V. S. Ecient optimal control of bioprocesses using second-order information.

Industrial & engineering chemistry

research 2000, 39, 42874295. (71) Tholudur, A.; Ramirez, W. F. Obtaining smoother singular arc policies using a modied iterative dynamic programming algorithm.

International Journal of Control 1997, 68,

11151128.

(72) Lee, J.; Ramirez, W. F. On-line optimal control of induced foreign protein production by recombinant bacteria in fed-batch reactors.

Chemical engineering science 1996, 51,

521534.

(73) Lin, Q.; Ryan, L.; Xu, C.; Lay, T. K. State-delay estimation for nonlinear systems using inexact output data. Control Conference (CCC), 2014 33rd Chinese. 2014; pp 65496554.

(74) Liu, C.; Gong, Z.; Teo, K. L. Robust parameter estimation for nonlinear multistage time-delay systems with noisy measurement data.

2018, 53, 353368.

34

ACS Paragon Plus Environment

Applied Mathematical Modelling

Page 35 of 36 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Industrial & Engineering Chemistry Research

(75) Mayne, D. Q.; Michalska, H. Adaptive receding horizon control for constrained nonlinear systems. Decision and Control, 1993., Proceedings of the 32nd IEEE Conference on. 1993; pp 12861291.

(76) Mayne, D. Q.; Rawlings, J. B.; Rao, C. V.; Scokaert, P. O. Constrained model predictive control: Stability and optimality.

(77) Hamilton, J. D.

Automatica 2000, 36, 789814.

Time series analysis ; Princeton university press Princeton, 1994; Vol. 2.

(78) Cheein, F. A. A.; Pereira, F. M. L.; Di Sciascio, F.; Carelli, R. Autonomous simultaneous localization and mapping driven by Monte Carlo uncertainty maps-based navigation.

The Knowledge Engineering Review 2013, 28, 3557.

(79) Dochain, D.

Automatic control of bioprocesses ; John Wiley & Sons, 2013.

(80) Rani, K. Y.; Rao, V. R. Control of fermentersa review.

Bioprocess Engineering 1999,

21, 7788. (81) Tempo, R.; Ishii, H. Monte Carlo and Las Vegas Randomized Algorithms for Systems and Control*: An Introduction.

European journal of control 2007, 13, 189203.

(82) Wang, H.; Wang, C.; Wang, Y.; Gao, X.; Yu, C. Bayesian forecasting and uncertainty quantifying of stream ows using MetropolisHastings Markov Chain Monte Carlo algorithm.

Journal of Hydrology 2017, 549, 476483.

(83) Noruzi, A.; Banki, T.; Abedinia, O.; Ghadimi, N. A new method for probabilistic assessments in power systems, combining monte carlo and stochastic-algebraic methods.

Complexity 2015, 21, 100110.

35

ACS Paragon Plus Environment

Industrial & Engineering Chemistry Research 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Graphical TOC Entry

36

ACS Paragon Plus Environment

Page 36 of 36